suggest .. for erroneous ... in struct pattern; don't discard successfully-parsed fields on error
#46721
Conversation
Previously, if an error occured while parsing field patterns, we would just return the error, discarding any field patterns that did successfully parse, which sometimes resulted in spurious diagnostics. At the cost of a little boilerplate (the repetitious `.map_err(|err| ((fields.clone(), etc), err))` is annoying and the present author is sorry to say she could not do better), we can pack the successfully-parsed field patterns into the error and use them. This was inspired by rust-lang#46718.
kennytm
added
the
S-waiting-on-review
|
I wonder if it wouldn't be better to not print a message about missing field at all here. Using |
| let lo = self.span; | ||
| let hi; | ||
|
|
||
| if self.check(&token::DotDot) { | ||
| self.bump(); | ||
| if self.token != token::CloseDelim(token::Brace) { | ||
| let token_str = self.this_token_to_string(); | ||
| return Err(self.fatal(&format!("expected `{}`, found `{}`", "}", | ||
| token_str))) | ||
| let err = self.fatal(&format!("expected `{}`, found `{}`", "}", |
There was a problem hiding this comment.
format!("expected {}...", "}", ...) is a bit silly, could you change it? :)
| --> $DIR/issue-46718-expected-dotdot-found-dotdotdot.rs:21:13 | ||
| | | ||
| 21 | PersonalityInventory { expressivity: exp, ... } => exp | ||
| | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ missing field `instrumentality` |
There was a problem hiding this comment.
I agree with @xfix's comment, it'd be nice if this error wasn't shown. IIRC, this error happens during type check, so you'd have to modify the ast::StructPatKind and its equivalent in the hir to include a boolean recovered and only emit this error if it's true (I'm doing the same for type errors involving tail expressions and recovered if blocks in #46732). I'm sure that there're are other errors that could benefit from this signal too, but I wouldn't go looking to fix them on this PR. This change would probably cause many more files to be modified, so we could land this first and have a follow up PR if you prefer.
| // accept trailing commas | ||
| if self.check(&token::CloseDelim(token::Brace)) { break } | ||
| } | ||
|
|
||
| let attrs = self.parse_outer_attributes()?; | ||
| let attrs = self.parse_outer_attributes() | ||
| .map_err(|err| ((fields.clone(), etc), err))?; | ||
| let lo = self.span; | ||
| let hi; | ||
|
|
||
| if self.check(&token::DotDot) { |
There was a problem hiding this comment.
The only thing this PR needs to do is to replace this self.check(&token::DotDot) with self.check(&token::DotDot) || self.token == token::DotDotDot, and report a non-fatal error (self.span_err(...)) if it's actually a DotDotDot.
There was a problem hiding this comment.
I see ...
(I might claim that the extra machinery in this PR is still potentially useful for taking successfully-parsed field patterns into account when there's some other kind of syntax error ... but maybe that can be a separate, future PR if I find this PartialPResult device being useful elsewhere.)
|
Closing in favor of simpler alternative #46763; thanks to @petrochenkov for the suggestion. (The |
…_pattern_ellipsis, r=petrochenkov in which `..` is suggested for erroneous `...` in struct field patterns Resolves #46718. Supersedes #46721. r? @petrochenkov
Regarding #46718.
r? @pnkfelix